Skip to main content
Caronte
Docs

Admin guide · Platform

Monitoring

The landing page after login answers “what is my organization spending, and on what?”. Everything on it reflects real gateway traffic, aggregated live.

KPIs & charts

The dashboard combines headline numbers with trend and distribution views.

The Monitoring dashboard: the Performance and Compliance tabs, a KPI row, a project view stacked by project, top models used, and a model view.
Monitoring — the org-wide landing dashboard.
  • KPI row — total spend, tokens used, requests, and average cost per 1M tokens for the selected period.
  • Project view — cost (or tokens, via the toggle) per period as stacked bars, one segment per project. Hover a segment to follow that project across the range, or a column to light up the whole period.
  • Top models used — ranked by requests, with share and cost per model.
  • Model view and Provider → Models — where the volume actually flows: spend or tokens per period stacked by model, and the flow per provider.

tip · Export report

The button top-right renders the current view — filters included — as a report you can send on, rather than a screenshot of the page.

The Compliance tab

The same period, read as EU AI Act drift instead of spend: calls evaluated, violations, judge failures and cost, plus the declared-versus-deduced risk mix.

The Compliance tab: calls evaluated, violations, judge failures and judge cost over a guardrail activity chart and risk-level breakdowns.
Monitoring → Compliance — only meaningful once keys declare a risk level.

It is populated by the EU AI Act classifier; keys with no declaration are counted as skipped rather than compliant.

Slicing the data

The left rail narrows every widget at once — filters compose.

  • Period — the date range all KPIs and charts are computed over.
  • Project — restrict to one or more projects.
  • Environment / Tenant — the metadata carried by API keys and requests.

note · Visibility is role-scoped

Org owner / admin see the whole organization. A plain member only sees data for the projects they are on the roster of.

Drilling into traces

Aggregates link down to individual requests. Each project has a traces view — click its row in the Projects list — listing every trace with latency and cost. Click a row for the generation timeline with inputs and outputs.

tip · Make traces readable from code

Clients can name traces and group turns with trace_id, session_id and friends — see the developer page Monitoring & traces.